I understand the problem - sampling for a value of t over a curve between 0 and 1 does not necessarily give you an even spatial distance over the curve, which is a natural consequence of the way the curve formula works.
I've been researching some other alternatives. Catmull Rom is supposed to give a constant speed but requires some really weird control points that would be counter-intuitive in the editor. Trigonometric splines are of interest, but again I can't quite work out how well that's going to work in terms of UI.
Bugger. Thought that quadratic splines would work - they provide the nicest concept from a UI point of view.
I have a ridiculous idea that I could have some kind of mapping between 0 and 1 onto a section of the curve that took into account the linear distance between the last point and the desired new point, binary chopping between different values of t until a point on the curve was found that satisfied a tolerance of the actual distance the point moved, but that's probably just mental. Hard to tell.
Blibble blibble. Is it actually that important to be able to sweep the camera in smooth curves around the level at the start? Probably not.
With the integral in hand, you could then evaluate the integral over different ranges of the input parameter and see how long it was. This makes it simple to create a mapping between linear [0,1] range and a distance based [0,1] range that is distorted to the appropriate distance metric.
Its worth a shot if you don't find any other alternatives...